tipc-discussion

[tipc-discussion] [net] tipc: missing entries in name table of publications

[Hoang Le <hoa...@de...>]

When binding multiple services with specific type 1Ki, 2Ki..,
this leads to some entries in the name table of publications
missing when listed out via 'tipc name show'.

The problem is at identify zero last_type conditional provided
via netlink. The first is initial 'type' when starting name table
dummping. The second is continuously with zero type (node state
service type). Then, lookup function failure to finding node state
service type next iteration.

To solve this, adding more conditional to marked as dirty type and
lookup correct service type for the next iteration instead of select
the first service as initial 'type' zero.

Signed-off-by: Hoang Le <hoa...@de...>
---
 net/tipc/name_table.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/tipc/name_table.c b/net/tipc/name_table.c
index bff241f03525..89993afe0fbd 100644
--- a/net/tipc/name_table.c
+++ b/net/tipc/name_table.c
@@ -909,7 +909,8 @@ static int tipc_nl_service_list(struct net *net, struct tipc_nl_msg *msg,
 	for (; i < TIPC_NAMETBL_SIZE; i++) {
 		head = &tn->nametbl->services[i];
 
-		if (*last_type) {
+		if (*last_type ||
+		    (!i && *last_key && (*last_lower == *last_key))) {
 			service = tipc_service_find(net, *last_type);
 			if (!service)
 				return -EPIPE;
-- 
2.17.1

[tipc-discussion] [net] tipc: missing entries in name table of publications

[Hoang Le <hoa...@de...>]

When binding multiple services with specific type 1Ki, 2Ki..,
this leads to some entries in the name table of publications
missing when listed out via 'tipc name show'.

The problem is at identify zero last_type conditional provided
via netlink. The first is initial 'type' when starting name table
dummping. The second is continuously with zero type (node state
service type). Then, lookup function failure to finding node state
service type in next iteration.

To solve this, adding more conditional to marked as dirty type and
lookup correct service type for the next iteration instead of select
the first service as initial 'type' zero.

Acked-by: Jon Maloy <jon...@er...>
Signed-off-by: Hoang Le <hoa...@de...>
---
 net/tipc/name_table.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/tipc/name_table.c b/net/tipc/name_table.c
index bff241f03525..89993afe0fbd 100644
--- a/net/tipc/name_table.c
+++ b/net/tipc/name_table.c
@@ -909,7 +909,8 @@ static int tipc_nl_service_list(struct net *net, struct tipc_nl_msg *msg,
 	for (; i < TIPC_NAMETBL_SIZE; i++) {
 		head = &tn->nametbl->services[i];
 
-		if (*last_type) {
+		if (*last_type ||
+		    (!i && *last_key && (*last_lower == *last_key))) {
 			service = tipc_service_find(net, *last_type);
 			if (!service)
 				return -EPIPE;
-- 
2.17.1

Re: [tipc-discussion] [net] tipc: missing entries in name table of publications

[David M. <da...@da...>]

From: Hoang Le <hoa...@de...>
Date: Tue,  9 Apr 2019 14:59:24 +0700

> When binding multiple services with specific type 1Ki, 2Ki..,
> this leads to some entries in the name table of publications
> missing when listed out via 'tipc name show'.
> 
> The problem is at identify zero last_type conditional provided
> via netlink. The first is initial 'type' when starting name table
> dummping. The second is continuously with zero type (node state
> service type). Then, lookup function failure to finding node state
> service type in next iteration.
> 
> To solve this, adding more conditional to marked as dirty type and
> lookup correct service type for the next iteration instead of select
> the first service as initial 'type' zero.
> 
> Acked-by: Jon Maloy <jon...@er...>
> Signed-off-by: Hoang Le <hoa...@de...>

Applied, and queued up for -stable.

Please provide an appropriate Fixes: tag next time.

Thank you.

Re: [tipc-discussion] [net] tipc: missing entries in name table of publications

[Jon M. <jon...@er...>]

Nice job. I am just flabbergasted by the fact that this seems to be a bug that has always been around, and still it wasn't seen until now, with all the systems with huge name tables we have running around in the world.
Maybe most people are still using "tipc-config"....

Acked-by: Jon Maloy <jon...@er...>


> -----Original Message-----
> From: Hoang Le <hoa...@de...>
> Sent: 8-Apr-19 07:08
> To: Jon Maloy <jon...@er...>; ma...@do...;
> yin...@wi...; tip...@li...
> Subject: [net] tipc: missing entries in name table of publications
> 
> When binding multiple services with specific type 1Ki, 2Ki.., this leads to some
> entries in the name table of publications missing when listed out via 'tipc
> name show'.
> 
> The problem is at identify zero last_type conditional provided via netlink. The
> first is initial 'type' when starting name table dummping. The second is
> continuously with zero type (node state service type). Then, lookup function
> failure to finding node state service type next iteration.
> 
> To solve this, adding more conditional to marked as dirty type and lookup
> correct service type for the next iteration instead of select the first service as
> initial 'type' zero.
> 
> Signed-off-by: Hoang Le <hoa...@de...>
> ---
>  net/tipc/name_table.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/net/tipc/name_table.c b/net/tipc/name_table.c index
> bff241f03525..89993afe0fbd 100644
> --- a/net/tipc/name_table.c
> +++ b/net/tipc/name_table.c
> @@ -909,7 +909,8 @@ static int tipc_nl_service_list(struct net *net, struct
> tipc_nl_msg *msg,
>  	for (; i < TIPC_NAMETBL_SIZE; i++) {
>  		head = &tn->nametbl->services[i];
> 
> -		if (*last_type) {
> +		if (*last_type ||
> +		    (!i && *last_key && (*last_lower == *last_key))) {
>  			service = tipc_service_find(net, *last_type);
>  			if (!service)
>  				return -EPIPE;
> --
> 2.17.1

Re: [tipc-discussion] [net] tipc: missing entries in name table of publications

[Hoang L. <hoa...@de...>]

Hi Jon,

In my reproduction, I'm seeing the problem either tipc-config. 
I guest almost systems were not using as special service type as my test, then all node state service type fit in one skb. 

Regards,
Hoang
-----Original Message-----
From: Jon Maloy <jon...@er...> 
Sent: Tuesday, April 9, 2019 5:39 AM
To: Hoang Huu Le <hoa...@de...>; ma...@do...; yin...@wi...; tip...@li...
Subject: RE: [net] tipc: missing entries in name table of publications

Nice job. I am just flabbergasted by the fact that this seems to be a bug that has always been around, and still it wasn't seen
until now, with all the systems with huge name tables we have running around in the world.
Maybe most people are still using "tipc-config"....

Acked-by: Jon Maloy <jon...@er...>


> -----Original Message-----
> From: Hoang Le <hoa...@de...>
> Sent: 8-Apr-19 07:08
> To: Jon Maloy <jon...@er...>; ma...@do...;
> yin...@wi...; tip...@li...
> Subject: [net] tipc: missing entries in name table of publications
> 
> When binding multiple services with specific type 1Ki, 2Ki.., this leads to some
> entries in the name table of publications missing when listed out via 'tipc
> name show'.
> 
> The problem is at identify zero last_type conditional provided via netlink. The
> first is initial 'type' when starting name table dummping. The second is
> continuously with zero type (node state service type). Then, lookup function
> failure to finding node state service type next iteration.
> 
> To solve this, adding more conditional to marked as dirty type and lookup
> correct service type for the next iteration instead of select the first service as
> initial 'type' zero.
> 
> Signed-off-by: Hoang Le <hoa...@de...>
> ---
>  net/tipc/name_table.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/net/tipc/name_table.c b/net/tipc/name_table.c index
> bff241f03525..89993afe0fbd 100644
> --- a/net/tipc/name_table.c
> +++ b/net/tipc/name_table.c
> @@ -909,7 +909,8 @@ static int tipc_nl_service_list(struct net *net, struct
> tipc_nl_msg *msg,
>  	for (; i < TIPC_NAMETBL_SIZE; i++) {
>  		head = &tn->nametbl->services[i];
> 
> -		if (*last_type) {
> +		if (*last_type ||
> +		    (!i && *last_key && (*last_lower == *last_key))) {
>  			service = tipc_service_find(net, *last_type);
>  			if (!service)
>  				return -EPIPE;
> --
> 2.17.1

Re: [tipc-discussion] [net] tipc: missing entries in name table of publications

[Jon M. <jon...@er...>]

Hi Hoang,
Yes, but the code for collecting data in the stack was different with "native" tipc-config vs. now, when tipc-config in reality is using the same code as "tipc" just via the "compat" layer.
Not sure when this changed was done, but probably at the uplift to sp2.

///jon


> -----Original Message-----
> From: Hoang Le <hoa...@de...>
> Sent: 9-Apr-19 04:14
> To: Jon Maloy <jon...@er...>; ma...@do...;
> yin...@wi...; tip...@li...
> Subject: RE: [net] tipc: missing entries in name table of publications
> 
> Hi Jon,
> 
> In my reproduction, I'm seeing the problem either tipc-config.
> I guest almost systems were not using as special service type as my test, then
> all node state service type fit in one skb.
> 
> Regards,
> Hoang
> -----Original Message-----
> From: Jon Maloy <jon...@er...>
> Sent: Tuesday, April 9, 2019 5:39 AM
> To: Hoang Huu Le <hoa...@de...>; ma...@do...;
> yin...@wi...; tip...@li...
> Subject: RE: [net] tipc: missing entries in name table of publications
> 
> Nice job. I am just flabbergasted by the fact that this seems to be a bug that
> has always been around, and still it wasn't seen until now, with all the
> systems with huge name tables we have running around in the world.
> Maybe most people are still using "tipc-config"....
> 
> Acked-by: Jon Maloy <jon...@er...>
> 
> 
> > -----Original Message-----
> > From: Hoang Le <hoa...@de...>
> > Sent: 8-Apr-19 07:08
> > To: Jon Maloy <jon...@er...>; ma...@do...;
> > yin...@wi...; tip...@li...
> > Subject: [net] tipc: missing entries in name table of publications
> >
> > When binding multiple services with specific type 1Ki, 2Ki.., this
> > leads to some entries in the name table of publications missing when
> > listed out via 'tipc name show'.
> >
> > The problem is at identify zero last_type conditional provided via
> > netlink. The first is initial 'type' when starting name table
> > dummping. The second is continuously with zero type (node state
> > service type). Then, lookup function failure to finding node state service
> type next iteration.
> >
> > To solve this, adding more conditional to marked as dirty type and
> > lookup correct service type for the next iteration instead of select
> > the first service as initial 'type' zero.
> >
> > Signed-off-by: Hoang Le <hoa...@de...>
> > ---
> >  net/tipc/name_table.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/net/tipc/name_table.c b/net/tipc/name_table.c index
> > bff241f03525..89993afe0fbd 100644
> > --- a/net/tipc/name_table.c
> > +++ b/net/tipc/name_table.c
> > @@ -909,7 +909,8 @@ static int tipc_nl_service_list(struct net *net,
> > struct tipc_nl_msg *msg,
> >  	for (; i < TIPC_NAMETBL_SIZE; i++) {
> >  		head = &tn->nametbl->services[i];
> >
> > -		if (*last_type) {
> > +		if (*last_type ||
> > +		    (!i && *last_key && (*last_lower == *last_key))) {
> >  			service = tipc_service_find(net, *last_type);
> >  			if (!service)
> >  				return -EPIPE;
> > --
> > 2.17.1
>